The Global Polycrisis and the Architecture of Systemic Risk

The diagram summarises the register. Thin grey lines join domains whose shocks strike the same country in the same year more often than chance. Arrows mark next-year transmission that remains statistically clear after accounting for income: a shock in one domain this year followed by a shock in another next year, among countries not already in shock there. Three arrows survive, and all three point to demography: from the economy, from technology and from governance.

This structure matches what humanitarian systems see on the ground. Recessions, currency collapses and inflation erode incomes and food security, and in fragile states they precede displacement and excess mortality. Internet shutdowns are usually ordered during unrest, elections or military operations, so they act as an early signal of the repression and violence that drive people from their homes. Conflict escalation and autocratisation push people across borders and into camps.

Climate stands somewhat apart in this annual, national view. Climate shocks co-occur with demographic shocks (lift 1.8) and governance shocks (1.4), but they do not significantly predict next-year shocks in other domains; if anything, economic shocks are less common in the year after a disaster. This does not mean climate risk is contained. Its effects on food, health, migration and conflict often build over several years and act locally, below the national thresholds used here, and the register cannot see the slow-onset stresses that climate change adds.

Using the four domains observed throughout (climate, economy, demography and governance), the share of countries with shocks in two or more domains in a year averaged 5.8% between 2000 and 2019. Peaks came with the global financial crisis (8.9% in 2008) and with regional wars and epidemics. In 2020, the pandemic pushed it to 32.7%, as recessions and falls in life expectancy struck almost everywhere at once. Since then it has stayed elevated, averaging 9.2% in 2021–2024. Triple-domain years rose from 1.0% to 3.7% of countries.

The domain lines show why. Economic shocks spiked in 2009 (36% of countries) and 2020 (64%), the two global synchronisers. Demographic shocks reached 39% in 2020 and 38% in 2021 as COVID cut life expectancy, then fell back to 7–11% a year as displacement from wars in Ukraine, Sudan and elsewhere replaced pandemic mortality. Governance shocks have run at 8–12% of countries a year throughout, with autocratisation onsets adding to conflict escalations. Internet shutdowns rose from 19% of countries in 2019 to 25% in 2024.

Including technology from 2016 adds to the count: with five domains, 14–21% of countries a year had multi-domain shocks in 2021–2024. The picture is consistent with the polycrisis narrative, though more specific. The world did not tip into permanent multi-domain crisis. It experienced one great synchronised shock and has since settled at a higher baseline, concentrated in a group of persistently fragile states.

Co-occurrence lift measures how much more often two shocks strike the same country in the same year than if they were independent. A lift of 1 means no association, 2 means twice as often as chance. Governance and demography have the highest lift (2.5), reflecting how conflict and political crises displace people almost immediately. Economy and demography follow (2.3), then climate and demography and technology and demography (1.8 each), economy and governance and governance and technology (1.6 each).

Demography appears in four of the six strongest pairs. It acts less as an independent source of shocks than as the domain where other shocks show up as human consequences: people fleeing, people dying early. This is the most important structural feature of the observed architecture. It means the human toll of polycrisis is concentrated in one place, where it can be anticipated and met.

The weakest pairs involve climate. Climate and economy co-occur only slightly more than chance (1.08), and climate and technology barely at all (1.10). At the national, annual level, most disasters do not coincide with recessions or currency crises, partly because large economies absorb them and partly because the economic damage of disasters is spread over years. The coupling of climate with the rest of the system is real in the literature but weaker in this kind of register than the other couplings.

Transmission asks whether a shock in one domain this year makes a shock in another domain more likely next year, among countries not already in shock in that second domain. Comparisons are made within income groups and pooled, so that poverty, which raises the risk of everything, does not create spurious links. Of the 20 possible directed links, three have risk ratios whose 95% intervals exclude 1.

Economic shocks are followed by demographic shocks 12.7% of the time, against 4.6% without one, an income-adjusted risk ratio of 2.5 (1.8–3.3). Internet shutdowns are followed by demographic shocks 19.8% of the time against 9.3% (2.0, 1.3–2.9), and governance shocks 10.1% against 5.1% (1.6, 1.1–2.4). Several other links point the same way but are not statistically clear: technology to governance (1.3), governance to economy (1.2) and economy to governance (1.2).

One link runs the other way. Countries with a climate shock are less likely to suffer an economic shock the next year (risk ratio 0.65), perhaps because reconstruction spending lifts measured growth or because the most disaster-prone countries have economies less exposed to global finance. The forest plot shows every link with its uncertainty. The few strong links are the ones that early warning and prevention should target.

Cascades extend over more than one year. Taking every new shock (no shock in that domain in the previous two years) and looking three years ahead, 53% of new internet-shutdown episodes were followed by shocks in at least two other domains, 44% of new demographic shocks, 34% of governance shocks, 32% of economic shocks and 29% of climate shocks. Shocks also persist within domains: a shutdown this year makes another next year five times more likely (55% against 11%), a demographic shock eight times (42% against 5.5%), and climate and economic shocks four times.

The burden falls unevenly. Since 2000, low-income countries have had an average of 5.8 years with shocks in two or more domains, lower-middle-income countries 3.0, upper-middle-income 2.3 and high-income 0.5. For three or more domains the averages are 1.8 against 0.04. Twenty-eight countries with 1.1 billion people had five or more multi-domain years, and 42 had none.

At the top are the long-running crises: Afghanistan (15 multi-domain years), Syria and Somalia (12), Yemen (11), and Sudan, Zimbabwe, Myanmar and the Central African Republic (9). Syria had seven years with shocks in three or more domains. These are not countries hit by an unlucky series of independent events. They are systems in which shocks feed each other and the buffers have been worn away.

A simple early-warning model uses the register's own regularities. For each domain, a logistic regression predicts next year's shock from this year's shocks in all five domains and the country's income group. Applying it to each country's latest state (2025 where available, 2024 otherwise) gives the probability of shocks in each domain in 2026 and the probability of shocks in two or more.

The highest risks are in countries already in multi-domain shock. South Sudan (69%) and Sudan (61%) top the list, followed by the DR Congo (57%), Afghanistan and Chad (51%), Mozambique (48%), Iran (47%), Iraq and Niger (43%), and Myanmar and Venezuela (41%). Eighteen countries were in shock in two or more domains in their latest year, and three (Iran, Sudan and South Sudan) in three.

This is a structural model, not a forecast of specific events. It knows nothing about elections, ceasefires, harvests or commodity prices, and its accuracy is limited by thin data on rare events. Its value is in showing where the architecture is already loaded, where the next shock is most likely to cascade, and therefore where anticipatory finance, humanitarian preparedness and diplomatic attention are most needed.

The polycrisis layer counts years since 2000 with shocks in two or more domains. It is highest across the Sahel, the Horn of Africa, Central Africa, the Middle East and Afghanistan, with Venezuela and Haiti in the Americas and Myanmar in Asia. Most of Europe, North America and East Asia record none or one, mostly in 2020.

The domain layers count shock years in each domain. Climate shocks concentrate in South and South-East Asia, the Horn of Africa, southern Africa, Central America and the Caribbean. Economic shocks concentrate in commodity exporters and countries with recurrent inflation or currency crises. Shutdowns cluster in South Asia, the Middle East and parts of Africa. Demographic and governance shocks map the world's conflict zones.

The 2026 layer shows the modelled probability of shocks in two or more domains. Hover over or tap a country for its card, which shows its 26-year shock strip, domain counts and 2026 probabilities. Clicking a country loads its latest state into the simulator in section 8. The sortable table below lists all 160 countries.

Choose an income group and switch on the shocks a country is experiencing this year. The simulator applies the early-warning model to show the probability of a shock in each domain next year, and the probability of shocks in two or more. Choosing a country loads its latest state and income group.

The most striking pattern is persistence: switching on a domain mostly raises the chance of a shock in the same domain next year. The cross-domain effects follow the transmission map, with economic, technology and governance shocks raising the next year's demographic risk. Income group matters greatly: the same shocks imply much higher multi-domain risk in a low-income country than in a high-income one.

The model is fitted to 2016–2024, the years in which all five domains are observed, so it reflects a period dominated by the pandemic and by new and renewed wars. Treat the outputs as a structured way to think about exposure to cascades, not as predictions.

Protect the sink. Because shocks across domains converge on displacement and early death, the highest-return investments are those that stop economic, political and climate shocks from becoming human catastrophes. These include social protection that scales up automatically when incomes collapse, anticipatory humanitarian finance triggered by forecasts, health systems that keep services running in crises, and early warning that reaches people at risk. They do not prevent the first shock, but they cut the most common path from shock to human loss.

Cut the strongest lines. Economic shocks are the most frequent source of transmission to demography. Macroeconomic stabilisers, debt standstills and emergency liquidity for vulnerable economies, together with food-price and fuel-price buffers, weaken that link. Conflict prevention and protecting democratic institutions weaken the governance link. Treat shutdowns as alarms. Internet shutdowns are an observable, near-real-time signal of the repression and unrest that precede displacement. Monitoring them belongs in early-warning systems, and resisting them protects the information people need to stay safe.

Watch the loaded nodes, and keep climate in view. The persistently fragile countries where several domains are already in shock carry most of the multi-domain risk. They need sustained, joined-up engagement across humanitarian, development and peace actors, not sequential crisis responses. Climate's coupling looks loose in an annual national register, but its effects accumulate over longer periods and at local scale, and the register undercounts heat. Mitigation and adaptation reduce a source of shocks that will matter more over time.

Panel: 160 countries with at least one million people and a World Bank income group, 2000–2025 (climate and demography to 2024; technology 2016–2025). Climate: EM-DAT disaster deaths excluding extreme temperature ≥ 10 per million, or people affected ≥ 5% of population (via Our World in Data). Economy: NY.GDP.PCAP.KD.ZG ≤ −3; FP.CPI.TOTL.ZG ≥ 20, at least double and 10 points above its previous three-year mean; PA.NUS.FCRF up ≥ 30% year on year. Technology: any shutdown recorded by Access Now #KeepItOn (combined 2016–2025 sheet). Demography: SP.DYN.LE00.IN down ≥ 1 year, or UNHCR refugees + asylum seekers + IDPs by origin up ≥ 0.5% of population. Governance: UCDP GED organised-violence deaths ≥ 1,000 and doubling, ≥ 100 and tripling, or ≥ 25 after two years below 25; or V-Dem ERT v16 autocratisation-episode onset or autocratic regime transition.

Statistics: lift = P(A∧B) ÷ (P(A)P(B)) within country-years; transmission = Mantel–Haenszel risk ratio of B at t+1 given A at t, among country-years with B = 0 at t, stratified by income group, with 95% intervals (Greenland–Robins variance); persistence = P(A at t+1 | A at t) against P(A at t+1 | not A); cascade = among new shocks (none in that domain in the previous two years), share with shocks in two or more other domains in years t to t+2. Early-warning model: one logistic regression per domain on the five shock indicators at t and income dummies, 2016–2024 (about 1,270 observations each), applied to the latest state.

Caveats: thresholds are judgement calls; disaster, displacement and shutdown records are incomplete, and heat deaths are excluded; life-expectancy series are modelled; internet shutdowns overlap with governance; national annual data miss subnational and slow-onset dynamics; associations are not causal; rare events make model estimates imprecise. Perception comparison: WEF Global Risks Report 2026.